AI Application Layer Unicorn LiblibAI Secures $300 Million in B+ Round, Valued at Over $20 Billion

LiblibAI, a unicorn in the AI application layer and parent company of Yanyu Technology, has completed a B+ round of financing of nearly $300 million, achieving a valuation exceeding $2 billion—the largest single financing round in China’s AI application layer to date. The company began as an AI art community and has amassed over 30 million users and more than 500,000 original models, generating over 5 million images daily. By 2026, its annual recurring revenue (ARR) is projected to exceed $300 million, with year-over-year revenue growth surpassing 3,000%. Meanwhile, following the release of the GLM-5.2 model, Zhipu’s market capitalization has surpassed HK$1 trillion, in contrast to MiniMax’s stock price halving. Capital trends are shifting: investors are no longer paying for concepts and imagination, but are instead placing greater emphasis on a model’s actual capabilities and product commercialization performance.

Article author and source: Blue Letter Project

Capital, changed?

Another "largest funding in the sector" has been announced.

Recently, the largest single funding round for a domestic AI application-layer company—amounting to nearly $300 million—went to a company that originated from an AI art community.

On June 18, LiblibAI’s parent company, Yanyu Technology, announced the completion of its B+ round of financing, with a post-money valuation exceeding $2 billion. Granite Asia, Tencent, and Shunwei Capital jointly led the round, with HT Investment and Shidai Capital participating as follow-on investors. Existing shareholders including Sequoia Capital China, Gao榕 Capital, and Ant Group further increased their investments.

However, in our conventional perception, companies that secure hundreds of millions of dollars in funding at once are typically those developing large models. The investment list above also resembles capital’s concerted pursuit of a star large model company.

However, Yanyu Technology did not train its own general-purpose large model. Three years ago, its first project was a community that allowed creators to upload, train, and share AI art models.

In other words, while the most sought-after companies in the AI industry are still competing over parameters, computing power, and model capabilities, a company that doesn’t train foundational models but focuses solely on deploying them has quietly emerged as a dark horse.

Why is an AI art community worth $2 billion?

The key terms behind this surge in valuation and the successful $300 million funding round by Yuyu Technology are "AI application layer," but unlike traditional concepts such as large models, algorithms, and computing power, this term may seem less familiar.

So, to understand why Yanyu Technology is valued at $2 billion, you first need to understand what the AI application layer actually does.

In March this year, NVIDIA CEO Jensen Huang published a major long-form article systematically breaking down the five-layer architecture of the AI industry; however, even before Huang’s “lesson,” the AI industry had already adopted a simpler, more direct way of categorizing itself.

In simple terms, the AI industry can be roughly divided into three layers: computing power, models, and applications.

Model companies develop general capabilities that enable machines to understand language, generate images, and write code; application companies integrate these capabilities into specific products and embed them into workflows such as design, marketing, short-form video production, and e-commerce, ultimately identifying users and monetization opportunities.

Model capabilities determine what AI can do; the application layer determines who will use it, for how long, and whether they’re willing to pay continuously.

This categorization is also closely related to changes in the AI sector.

Over the past few years, general-purpose large models were scarce enough that capital naturally flowed first to model companies.

However, as open-source models and APIs become increasingly mature, the cost and barriers to accessing powerful models continue to decline. Model capabilities that were once accessible only to a few companies are now becoming foundational capabilities that application businesses can purchase, integrate, and replace.

Additionally, as developers have access to an increasing number of models, particularly in general use cases such as image and text generation and basic interactions, the experience differences between various models are narrowing.

For application companies, the underlying model is gradually shifting from a core asset that must be developed in-house to a foundational capability that can be purchased, integrated, and flexibly replaced.

This is precisely the core value of AI application layers: enterprises do not need to focus on developing their own foundational large models; their competitive advantage lies in packaging standardized model technologies into scenario-based products and industry solutions that end users are willing to pay for and repurchase.

Of course, merely integrating a model doesn't make a product competitive. A simple generation entry point can easily be replicated by competitors.

Truly valuable applications require model adaptation, workflow design, user engagement, content distribution, and monetization to turn a one-time trial into frequent usage.

The underlying model can be replaced, but the accumulated user relationships, industry data, and production processes are difficult to replicate in a short time.

Therefore, user base, retention, and revenue have become the key metrics by which capital evaluates Liblib; meanwhile, the creator community, model and asset resources, distribution channels, and workflows constitute the truly irreplicable core of its advantage.

In 2023, Stable Diffusion sparked the popularity of AI-generated art, leading to a surge in domestic image-generation websites. With similar models and comparable features, the cost for users to switch platforms is low.

Liblib chose to start by bringing together creators who can train models, tune parameters, and build workflows.

Founder Chen Mian previously oversaw the monetization of Jianying and CapCut and is no stranger to creator tools. After LiblibAI’s launch, creators can upload, train, and share LoRA models and workflows on the platform, while regular users can directly utilize these resources to complete their creations.

The more users on the platform, the richer the models and assets become; and the richer the content, the more creators are attracted to stay. What Liblib has accumulated is not just a generation tool, but a self-expanding creator ecosystem.

According to the company, LiblibAI has accumulated over 30 million users, more than 500,000 original models, and over 100 million professional images and video assets, generating more than 5 million images daily.

The real importance of these numbers lies in their ability to be reused.

The AI-designed agent Xingliu advances AI capabilities from generating single images to delivering complete designs; launching in early 2026, LibTV further enters the production of short dramas, films, and brand videos.

After launching the new product, Yuyu Tech did not need to seek its first users from scratch—the existing community already provided creators, models, assets, and distribution channels.

The company disclosed that, within the first month of LibTV's launch, daily revenue exceeded $1 million; by May, monthly revenue had surpassed 13 times that of the launch month, serving nearly a thousand short-form drama teams, film and television production companies, advertising agencies, and brand clients.

The AI industry lacks a product that suddenly goes viral; the real challenge lies in replicating one success to the next. Yanyu Technology has truly differentiated itself by moving from LiblibAI to Xingliu, and then to LibTV—its real strength is consistently identifying user needs and leveraging its existing ecosystem to launch new businesses.

As of May 2026, Yanyu Technology disclosed that its ARR (Annualized Recurring Revenue) exceeded $300 million, with the group's total revenue growing by more than 3,000% year-over-year.

Thus, capital is drawn to the AI art community’s ability to consistently turn new technologies into products, and those products into revenue.

But more interesting than the amount Liblib raised is the fact that securing this largest funding in the sector signals a shift in capital trends.

The recent stock performance of Zhipu and MiniMax has provided the answer.

In March this year, MiniMax's stock price once surged to HK$1,330, with a market capitalization approaching HK$390 billion. However, since reaching its peak, the stock price has steadily declined, falling by half to its closing price on June 22.

On the other hand, Zhipu has shown a completely different price movement.

In mid-June, after Zhipu released GLM-5.2, its stock price rose rapidly. On June 22, its intraday market capitalization surpassed HK$1 trillion for the first time, further widening the gap with MiniMax.

Among recently listed Chinese large model companies, one has dropped by half from its peak, while another has surged to a market capitalization of one trillion yuan.

Unlocking expectations, circulating supply, and market sentiment all amplify price volatility for both companies. However, beyond these factors, products remain the most direct catalyst for shifting market expectations.

After the release of LM-5.2, it quickly garnered substantial positive feedback on code generation and long-range tasks.

After personally testing it, Vercel CEO Guillermo Rauch openly admitted that he was “almost shocked” by GLM-5.2’s programming capabilities; on Code Arena, where over a million users worldwide participated in blind tests, GLM-5.2 ranked as the top available model globally.

Moreover, GLM-5.2 is not cheap.

According to the official API pricing from both parties, for inputs not exceeding 512K tokens, MiniMax M3 charges $0.3 per million input tokens and $1.2 per million output tokens; GLM-5.2 charges $1.4 and $4.4 respectively, making its input price approximately 4.7 times that of M3 and its output price approximately 3.7 times that of M3.

Thus, Zhipu has not won over the market with ultra-low pricing; yet, as long as its models are sufficiently powerful, developers are willing to use them, and investors are willing to pay a higher premium. In contrast, M3, which was released earlier than GLM-5.2 and at a much lower price, has not achieved the same stock performance.

Ultimately, low prices can attract users to try, but whether users stay and whether the market continues to support the product depends on the product’s own competitiveness.

In fact, Zhipu's stock price has been rising steadily, for the same reason that Liblib secured nearly $300 million in funding: with strength, tangible capabilities, and impressive achievements, you can attract investment.

For Zhipu, success is measured by model capabilities, real-world task performance, and developer reputation; for Liblib, success is reflected in over 30 million users, 500,000 original models, nearly a thousand professional content clients, and more than $300 million in ARR.

Capital is not choosing between large models and application layers; it is simply becoming less willing to pay for a concept or a blueprint filled with imagination.

The hotter AI becomes, the more debates arise around bubbles and valuations. These arguments are unlikely to be resolved in the short term, but capital has already begun filtering companies in a more direct way: whether the models actually work, whether anyone is using the products, and whether customers are willing to pay consistently—ultimately, results speak for themselves.

Although Liblib and Zhipu, among other companies, occupy different positions in the industrial chain, they have delivered the same result.

A trend can get a company to the table, but it’s performance that determines who ultimately keeps the money.

Blue Letter Program